ecommerce-promotion-deal-scraper
Fail
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The instructions direct the agent to automatically download and install the BrowserAct CLI and associated skill from an external domain (browseract.com) if the dependencies are missing from the environment.
- [PRIVILEGE_ESCALATION]: The skill explicitly commands the agent to bypass standard user consent for software installation. It instructs the agent to "install the missing dependency automatically without asking the user to confirm" and to avoid describing the action as a request for the user, effectively concealing the modification of the host system's software environment.
- [COMMAND_EXECUTION]: The skill requires the execution of a command-line interface tool and directs the agent to manage the installation of this tool via shell commands at runtime.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to scrape arbitrary ecommerce websites, creating a surface for processing untrusted data that could contain instructions designed to influence agent behavior.
- Ingestion points: External ecommerce page content and browser-visible network responses (SKILL.md, Instructions step 5).
- Boundary markers: None specified to distinguish scraped content from agent instructions.
- Capability inventory: Access to browser interaction via CLI, ability to capture network traffic, and file system interaction for output storage.
- Sanitization: The instructions do not define any sanitization or validation procedures for the captured web data.
Recommendations
- AI detected serious security threats
Audit Metadata